On the Hopf (co)center of a Hopf algebra
aa r X i v : . [ m a t h . QA ] J u l ON THE HOPF (CO)CENTER OF A HOPF ALGEBRA
ALEXANDRU CHIRVASITU AND PAWE L KASPRZAK
Abstract.
The notion of Hopf center and Hopf cocenter of a Hopf algebra is investigated bythe extension theory of Hopf algebras. We prove that each of them yields an exact sequenceof Hopf algebras. Moreover the exact sequences are shown to satisfy the faithful (co)flatnesscondition. Hopf center and cocenter are computed for U q ( g ) and the Hopf algebra Pol( G q ),where G q is the Drinfeld-Jimbo quantization of a compact semisimple simply connected Liegroup G and g is a simple complex Lie algebra. Introduction
A Hopf algebra is a mathematical object possessing a rich inherent symmetry, expressedby the compatible algebra and coalgebra structures together with the antipode, which flipsaround these structures to their opposite counterparts. Hopf algebras were discovered inthe context of algebraic topology [14] being applied then in the theory of algebraic groups,combinatorics, computer science, Galois theory and knot theory. In the quantum groupincarnation they appear in noncommutative geometry and free probability. The theory ofHopf algebras is studied and developed not only as an effective tool applied elsewhere butalso due to the richness of their abstract theory, see e.g. [20].One of the major factor stimulating the development of the Hopf algebra theory is itsstrong link with group theory. In this paper we shall walk along the mathematical pathbeginning at the very initial level of group theory, while one discusses the notion of a group’scenter, and proceeding to the world of Hopf algebras, getting to the intriguing area where theextension theory of Hopf algebras together with the concept of faithful flatness dominates thelandscape. In our journey we shall experience not only the conceptual entertainment but weshall also compute the classes of motivating and instructive examples.Let G be a group. The center of G Z ( G ) = { s ∈ G : ∀ t ∈ G st = ts } is a subgroup of G which is normal: for all t ∈ G and s ∈ Z ( G ) we have tst − = s ∈ Z ( G ).In particular the quotient space G / Z ( G ) is equipped with the group structure such that thequotient map π : G → G / Z ( G ) is a surjective group homomorphism. Then ker π = Z ( G ) andwe have the short exact sequence in the category of groups { e } → Z ( G ) → G → G / Z ( G ) → { e } (0.1)Let k be a field. A finite group G yields a pair of Hopf algebra triples • the group algebras triple: k [ Z ( G )] , k [ G ] , k [ G / Z ( G )] • the k -valued functions triple k ( G / Z ( G )) , k ( G ) , k ( Z ( G )) PK was partially supported by the Harmonia NCN grant 2012/06/M/ST1/00169. which are expected to be incorporated into the pair of Hopf algebras exact sequences. Theconcept of exact sequences in the category of Hopf algebras was introduced in [2], and indeedapplied in our case gives rise to exact sequences k → k [ Z ( G )] → k [ G ] → k [ G / Z ( G )] → kk → k ( G / Z ( G )) → k ( G ) → k ( Z ( G )) → k (0.2)Dropping the group context in our example and emphasizing the Hopf context, we would beled to the idea of Hopf center and cocenter. It was first (partially) discussed in [1] where anarbitrary Hopf algebra A was assigned with the left and right counterparts of (0.2) k → HZ ( A ) → AA → HC ( A ) → k (0.3)The main subject of our paper concerns the corresponding right and left missing parts of theexact sequences. We show the exact sequences k → HZ ( A ) → A → B → kk → C → A → HC ( A ) → k (0.4)indeed exist, providing also certain accurate and/or approximate description of the exactsequences ingredients. Our results treat also the faithful (co)flatness issue, i.e. we show that A is always faithfully flat over HZ ( A ) ֒ → A and faithfully coflat over A ։ HC ( A ). This makesthe exact sequences (0.4) particularly nice.Passing to the examples we compute (co)centers of the quantized universal envelopingalgebra U q ( g ) and the quantized Hopf algebra Pol( G q ) of function on G q , the Drinfeld-Jimboquantization of a compact semisimple simply connected Lie group G with Lie algebra g . Acknowledgements.
We would like to thank the anonymous referee for many valuablesuggestions on improving the initial draft.1.
Preliminaries
Throughout, we work over an algebraically closed field k .Our main references for the general theory of Hopf algebras are [20, 23]. For a Hopf algebra A over k we denote ∆, ε , S respectively comultiplication counit and antipode of A . In order todistinguish coproducts of different Hopf algebras we shall write ∆ A and similarly S A and ε A for antipode and counit respectively. The kernel of counit will be denoted A + and for B ⊂ A we write B + = B ∩ A + . The Sweedler notation ∆( x ) = x (1) ⊗ x (2) , x ∈ A will be freely usedwhen convenient. The center of the algebra A will be denoted Z ( A ).The fact that we are working over an algebraically closed field ensures that the simplecoalgebras are exactly those dual to matrix algebras M n ( k ); we refer to such an object asa matrix coalgebra . The coradical of a coalgebra (i.e. the sum of its simple subcoalgebras;see [23, Section 9.0]) is a direct sum of matrix subcoalgebras. In particular, a cosemisimplecocommutative coalgebra will automatically be a group coalgebra; this remark is dual to thefact that the only semisimple commutative algebras over k are the finite products of copiesof k .The monoidal category of right A -comodules is denoted by M A , while the full subcategory offinite dimensional comodules will be denoted by M A f . Similarly, the category of left comodulesand the full subcategory of finite dimensional left comodules will be denoted by A M and A M f respectively. We will work mostly with right comodules. N THE HOPF (CO)CENTER OF A HOPF ALGEBRA 3
Recall that a monoidal category is left rigid if for every object x there is an object x ∗ (the left dual of x ) with morphisms ev : x ∗ ⊗ x → → x ⊗ x ∗ (where 1 is the monoidalunit) so that both x x ⊗ x ∗ ⊗ x x coev ⊗ id id ⊗ evidand x ∗ x ∗ ⊗ x ⊗ x ∗ x ∗ id ⊗ coev ev ⊗ ididcommute.Similarly, a monoidal category is right rigid if for every object x there is an object ∗ x (its right dual ) with morphisms ev : x ⊗ ∗ x → → ∗ x ⊗ x that make x a left dual to ∗ x .For any Hopf algebra A the category M A f is left rigid: for a finite comodule V the antipode S can be used to put the comodule structure f f (0) ⊗ f (1) such that f (0) ( v ) f (1) = f ( v (0) ) Sv (1) , ∀ v ∈ V , f ∈ V ∗ , (1.1)where v v (0) ⊗ v (1) is our version of Sweedler notation for right comodule structures. Thiscomodule structure is such that the usual evaluation V ∗ ⊗ V → k and dual basis coevaluationmap k → V ⊗ V ∗ are comodule morphisms.We can also make M A f into a right rigid monoidal category, but we need A to have bijectiveantipode: the procedure is the same as above, except that the inverse of the antipode is usedinstead of S in (1.1).The adjoint action Ad : A → End( A ) of a Hopf algebra on itself is Ad x y = x (1) yS ( x (2) ).The map Ad : A → End( A ) is an algebra homomorphism Ad xy = Ad x Ad y satisfying the Leibniz rule Ad x ( yz ) = Ad x (1) ( y ) Ad x (2) ( z )and ∆( Ad x ( y )) = x (1) y (1) S ( x (4) ) ⊗ Ad x (2) ( y (2) ) (1.2)It is known that Z ( A ) = { x ∈ A : Ad y ( x ) = ε ( y ) x for all y ∈ A } (1.3)For the proofs of the claims in the present paragraph see e.g. [15, Section 1.3].The adjoint coaction ad : A → A ⊗ A is ad ( x ) = x (2) ⊗ S ( x (1) ) x (3) (1.4)Then ad is a linear map satisfying [1, 2.3.1-2.3.3]. Moreover( ad ⊗ id)( ad ( x )) = x (3) ⊗ S ( x (2) ) x (4) ⊗ S ( x (1) ) x (5) = (id ⊗ ∆)( ad ( x )) (1.5)for all x ∈ A .We recall the definition of exact sequence of Hopf algebras following [2]. ALEXANDRU CHIRVASITU AND PAWE L KASPRZAK
Definition 1.1.
Consider a sequence of morphisms of Hopf algebras k → A ι −→ C π −→ B → k (1.6)We say that (1.6) is exact if(1) ι is injective;(2) π is surjective;(3) ker π = C ι ( A ) + ;(4) ι ( A ) = { x ∈ C : ( π ⊗ id)∆( x ) = ⊗ x } . (cid:7) In the sequel we will make frequent use of the notions of faithfully (co)flat morphisms ofHopf algebras. For completeness we give the following definition.
Definition 1.2.
A left module V over an algebra A is flat if the functor − ⊗ A V on rightmodules preserves monomorphisms (i.e. injections). V is faithfully flat if it is flat and − ⊗ A V is also faithful. Right-handed notions are defined analogously for right modules.A morphism A → B of algebras is left (right) (faithfully) flat if B is (faithfully) flat as aleft (right) A -module.A right comodule V over a coalgebra C is coflat if the cotensor product functor V (cid:3) C − : C M →
Vect (defined dually to the tensor product; see [7, § V is faithfully coflat if it is coflat and V (cid:3) C − is also faithful. Once more,there are analogous left-handed notions for left comodules.A coalgebra morphism C → D is left (right) (faithfully) coflat if C is (faithfully) coflat as aleft (right) D -comodule. (cid:7) Remark 1.3.
It is shown in [25] that a D -comodule is faithfully coflat if and only if it isan injective cogenerator , in the sense that it is injective and all indecomposable injective C -comodules appear as summands in D .In particular, if D is cosemisimple and C → D is a surjective morphism of coalgebras thenit is (left and right) faithfully coflat. Indeed, our assumptions ensure that all D -comodulesare injective and the surjection C → D splits. The conclusion then follows from the fact that D itself is an injective cogenerator in both M D and D M (this observation is dual to the factthat an algebra is a projective generator in its category of either left or right modules). (cid:7) Remark 1.4.
Suppose that the antipode S C of C is bijective. Let A ι −→ C be a faithfullyflat morphism of Hopf algebras and suppose that ι is ad-invariant, i.e. Ad x ( ι ( A )) ⊂ ι ( A ) forall x ∈ C . Then conditions (1),(2),(3) imply condition (4) of Definition 1.1. Conversely let C π −→ B be a faithfully coflat morphism of Hopf algebras and suppose that π is normal in thesense of [2, Definition 1.1.5], i.e.( π ⊗ id)∆( x ) = ⊗ x ⇐⇒ (id ⊗ π )∆( x ) = x ⊗ for all x ∈ C . Then conditions (1),(2),(4) imply condition (3) of Definition 1.1. (cid:7) The leg numbering notation will be used throughout the paper, i.e. for x ∈ A ⊗ A we define x , x , x ∈ A ⊗ A ⊗ A , where e.g. x = ⊗ x . We shall also write x for the flippedversion of x : if x = a ⊗ b then x = b ⊗ a ⊗ and similarly for x , x . Leg numberingnotation is also used for linear maps. If T : A ⊗ A → A ⊗ A then id ⊗ T : A ⊗ A ⊗ A → A ⊗ A ⊗ A will be denoted by T , etc. We warn the reader not to confuse the Sweedler notation wherethe bracketed indices are used, with the unbracketed leg numbering notation. N THE HOPF (CO)CENTER OF A HOPF ALGEBRA 5 Hopf center
Let us begin with the following definition.
Definition 2.1. [1, Definition 2.2.3] Let A be a Hopf algebra. The Hopf center HZ ( A ) of A is the largest Hopf subalgebra of A contained in the center Z ( A ). (cid:7) Analyzing the reasoning which proceeds [1, Definition 2.2.3] we see that if S is bijectivethen we have HZ ( A ) = { x ∈ A : (id ⊗ ∆)(∆( x )) ∈ A ⊗ Z ( A ) ⊗ A } (2.1)In what follows we shall give a slightly weaker condition for x ∈ A to belong to HZ ( A ).In order to see the relation with [1, Section 2] let us note that applying ε to the third legof (id ⊗ ∆)(∆( x )) ∈ A ⊗ Z ( A ) ⊗ A we get ∆( x ) ∈ A ⊗ Z ( A ). Let us emphasize that theinvertibility assumption which enters the formulation of the next theorem can be dropped,by using the Tannakian description of the Hopf center which we shall give later. For a locallycompact quantum group version of the following result see [16]. Theorem 2.2.
Let A be a Hopf algebra with bijective antipode. Let us define M = { x ∈ A : ∆( x ) ∈ A ⊗ Z ( A ) } Then M = HZ ( A ) .Proof. Let us first note that if x ∈ M then x = ( ε ⊗ id)(∆( x )) ∈ Z ( A ) thus M ⊂ Z ( A ). Letus show that ∆( M ) ⊂ Z ( A ) ⊗ Z ( A ) (2.2)Using (1.3) it is enough to show that for y ∈ M we have( Ad x ⊗ id)(∆( y )) = ε ( x )∆( y )for all x ∈ A . We compute( Ad x ⊗ id)(∆( y )) = x (1) y (1) S ( x (2) ) ⊗ y (2) = x (1) y (1) S ( x (3) ) ⊗ ε ( x (2) ) y (2) = x (1) y (1) S ( x (4) ) ⊗ x (2) y (2) S ( x (3) )= ∆( Ad x ( y ))= ∆( ε ( x )( y ))= ε ( x )∆( y )where in the third equality we used that y (2) ∈ Z ( A ); in the fourth equality we used (1.2); inthe fifth equality we used y ∈ Z ( A ).Noting that for x ∈ M (id ⊗ ∆)(∆( x )) = (∆ ⊗ id)(∆( x )) ∈ A ⊗ A ⊗ Z ( A )we get ∆( M ) ⊂ A ⊗ M (2.3)The invertibility of S yields S ( Z ( A )) = Z ( A ) thus for all x ∈ M we get∆( S ( x )) = ( S ⊗ S )(∆ op ( x )) ∈ S ( Z ( A )) ⊗ S ( Z ( A )) = Z ( A ) ⊗ Z ( A )Using (2.2) and the bijectivity of S we conclude that S ( M ) = M . The S -invariance of M together with (2.3) implies that M is a Hopf subalgebra. ALEXANDRU CHIRVASITU AND PAWE L KASPRZAK If N ⊂ Z ( A ) is a Hopf subalgebra then∆( N ) ⊂ A ⊗ Z ( A )thus N ⊂ M . This means that M is the largest Hopf subalgebra of A , and hence coincideswith HZ ( A ). (cid:4) Lemma 2.3.
Let A be a Hopf algebra with bijective antipode. Then the center of A equals HZ ( A ) = { x ∈ Z ( A ) : ∆( x ) ∈ Z ( A ) ⊗ Z ( A ) } Proof.
Clearly if x ∈ Z ( A ) and ∆( x ) ∈ Z ( A ) ⊗ Z ( A ) then using Theorem 2.2 we get x ∈HZ ( A ). Conversely if x ∈ HZ ( A ) then x ∈ Z ( A ) and ∆( x ) ∈ Z ( A ) ⊗ Z ( A ). (cid:4) Lemma 2.4.
Let A be a Hopf algebra with bijective antipode and V a right A -comodule withthe corresponding map ρ : V → V ⊗ A satisfying ρ ( V ) ⊂ V ⊗ Z ( A ) Then ρ ( V ) ⊂ V ⊗ HZ ( A ) In particular V can be viewed as a right HZ ( A ) -comodule. The same holds for left comodules.Proof. Let us note that (id ⊗ ∆)( ρ ( V )) = ( ρ ⊗ id)( ρ ( V )) ⊂ ( ρ ⊗ id)( V ⊗ Z ( A )) ⊂ V ⊗ Z ( A ) ⊗ Z ( A )Using Lemma 2.3 we conclude that ρ ( V ) ⊂ V ⊗ HZ ( A ) and the rest is clear. (cid:4) Let us note we have Ad x ( HZ ( A )) = ε ( x ) HZ ( A ), thus the embedding ι : HZ ( A ) → A isad-invariant. Since HZ ( A ) is commutative, we can use [4, Proposition 3.12] to conclude that ι is a faithfuly flat morphism. Using Remark 1.4 we get the following theorem. Theorem 2.5.
Let A be a Hopf algebra with bijective antipode. Then the embedding ι : HZ ( A ) → A is faithfully flat and there exists a Hopf algebra B and a surjective morphism π : A → B of Hopf algebras such that the following sequence of Hopf algebras is exact k → HZ ( A ) → A → B → k A Tannakian description of the Hopf center.
According to the general philosophyof Tannaka reconstruction (as presented e.g. in [21]), the inclusion HZ ( A ) ⊆ A of Hopfalgebras can be described completely by identifying the full inclusion M HZ ( A ) f → M A f ofcategories of finite-dimensional comodules. We refer to loc. cit. for background on Tannakareconstruction for coalgebras, bialgebras and Hopf algebras.The following discussion applies to bialgebras, so throughout § A will be a fixed butarbitrary bialgebra; when we require it to be a Hopf algebra (or Hopf algebra with bijectiveantipode) we will say so explicitly. Definition 2.6. An A -comodule V ∈ M A f is central if the flip map τ : V ⊗ W → W ⊗ V , v ⊗ w w ⊗ v is a comodule morphism for all W ∈ M A f . (cid:7) N THE HOPF (CO)CENTER OF A HOPF ALGEBRA 7
Consider the full subcategory
C ⊆ M A f consisting of the central comodules. Before statingthe next result, recall also the following definition from [21] (see Definition 2.2.10 therein);here, a subquotient of a comodule is a quotient of a subcomodule (or equivalently, a subco-module of a quotient comodule). Definition 2.7.
A full subcategory C of an abelian category D is closed if all subquotientsin D of an object x ∈ C are again in C . (cid:7) Lemma 2.8. (a)
C ⊆ M A f is a closed monoidal subcategory. (b) If A is a Hopf algebra, then C is left rigid. (c) If in addition the antipode of A is bijective, then C is both left and right rigid.Proof. We leave part (a) to the reader, and instead focus on part (b). In other words, we willargue that if M A f admits left duals V V ∗ , then C is closed under taking such duals. Theargument for right duals in part (c) is completely analogous, so we will omit that as well.To check (b), note first that for any V ∈ C the left dual V ∗ and the right dual ∗ V areisomorphic through the canonical map. Now, the flip induces a natural isomorphism V ⊗ − ∼ = − ⊗ V of endofunctors of M A f , and hence also induces a natural isomorphism between theirright adjoints, ∗ V ⊗ − ∼ = V ∗ ⊗ − and − ⊗ V ∗ respectively. (cid:4) According to [21, Lemma 2.2.12] (or rather a monoidal version thereof), this implies that C is of the form M C f for some sub-bialgebra C ⊆ A . We will see below that in the case of Hopfalgebras C is precisely the Hopf center discussed above. First, we need Lemma 2.9.
Let V ∈ M A f . Then V ∈ C ⊂ M A f if and only if the coefficient coalgebra C ( V ) ⊂ A of V is contained in the center.Proof. If the property from Definition 2.6 holds for a specific comodule W we will say that V commutes with W .Let v i and w k be bases for V and W respectively, and denote their respective comodulestructures by v j v i ⊗ a ij and w ℓ w k ⊗ b kℓ , where repeated indices indicate summation. Then, V commutes with W if and only if thediagram V ⊗ W V ⊗ W ⊗ AW ⊗ V W ⊗ V ⊗ A τ τ ⊗ id A (2.4)is commutative (the horizontal arrows being the comodule structure maps of the two tensorproducts).Applying the two paths (right-down and down-right) in (2.4) to the element v j ⊗ w ℓ ∈ V ⊗ W and then reading off the coefficient of w k ⊗ v i ∈ W ⊗ V , we see that this is equivalent to a ij b kℓ = b kℓ a ij for all possible index choices. In other words, all elements of the coefficientcoalgebra C ( V ) must commute with all elements of C ( W ). The conclusion follows from thefact that as W is allowed to range over all finite-dimensional comodules, the resulting C ( W )jointly span all of A . (cid:4) ALEXANDRU CHIRVASITU AND PAWE L KASPRZAK
Consider now the sub-bialgebra C which has C ⊆ M A f as its comodule category. It isthe linear span of the coefficients of its comodules, so it must be contained in the center Z ( A ). On the other hand, any sub-bialgebra of A contained in Z ( A ) corresponds via Tannakareconstruction to some closed monoidal subcategory of M A f consisting of central comodules,which must then be contained in C . In conclusion, denoting again by HZ ( A ) the largestsub-bialgebra of A contained in the center Z ( A ) (for consistency, and also because we aremostly interested in Hopf algebras anyway) we get Theorem 2.10.
The inclusion
C → M A f of the category of central comodules is equivalent to M HZ f ⊆ M A f , where HZ = HZ ( A ) is the largest sub-bialgebra of A contained in the center Z of A .Moreover, if A is a Hopf algebra (or Hopf algebra with bijective antipode), then HZ is alsothe largest Hopf subalgebra (respectively Hopf subalgebra with bijective antipode) contained in Z (cid:4) Finally, we have the following alternative characterization of the Hopf subalgebra
HZ ⊆ A from Theorem 2.10. Theorem 2.11.
Let A be an arbitrary Hopf algebra. The largest Hopf subalgebra HZ ⊆ A contained in the center Z of A coincides with the set M = { x ∈ A : ∆( x ) ∈ Z ⊗ A } Proof.
Clearly, HZ is contained in M ; the interesting inclusion is the opposite one.Let x ∈ M . Since ((∆ ⊗ id) ◦ ∆)( x ) = ((id ⊗ ∆) ◦ ∆)( x ) ∈ Z ⊗ A ⊗ A , we have ∆( x ) ∈ M ⊗ A , and hence M can be regarded as a right A -comodule. We can moreovercheck that ∆( M ) ⊆ Z ⊗ Z just as in the proof of Theorem 2.2, and hence the coefficientsof M , as a right A -comodule, lie in Z . Let us emphasize that at this stage of the proof ofTheorem 2.2 we do not use invertibility of S .It now follows from Lemma 2.9 that M is central and hence must be an HZ -comodule.Finally, the conclusion follows by applying ε ⊗ id to ∆( x ) ∈ M ⊗ HZ for x ∈ M . (cid:4) Remark 2.12.
We could just as easily carried out the discussion for left comodules. Sincethe characterization of HZ as the largest sub-bialgebra contained in the center is blind tothis distinction, we could have substituted A ⊗ Z for Z ⊗ A in Theorem 2.11. Thus, we canavoid antipode bijectivity in Theorem 2.2. (cid:7) Hopf cocenter
In [1, Section 2] the following notion is considered:
Definition 3.1.
Let π : A → B be a morphism of Hopf algebras. We say that π is cocentral if (id ⊗ π ) ◦ ∆ = (id ⊗ π ) ◦ ∆ op (cid:7) In [1, Lemma 2.3.7] it was proved that there exists a quotient Hopf algebra HC ( A ) of A such that all cocentral morphisms from A factor uniquely through HC ( A ); moreover, HC ( A )is automatically cocommutative. Strictly speaking, in loc. cit. Hopf algebras are always N THE HOPF (CO)CENTER OF A HOPF ALGEBRA 9 assumed to have bijective antipode. Nevertheless, the existence of HC ( A ) for arbitrary Hopfalgebras follows from its existence for Hopf algebras with bijective antipode.Indeed, note that a cocentral morphism A → B factors through the universal involutivequotient A = A / ( S − id) of A . This means that the quotient A → A → HC ( A ) has theuniversal property defining HC ( A ).With this in place, we can now give Definition 3.2.
The cocenter HC ( A ) of a Hopf algebra A is the quotient Hopf algebra A →HC ( A ) that is universal among cocentral quotients of A . (cid:7) A similar concept for cosemisimple Hopf algebras with the assumption that HC ( A ) iscosemisimple was introduced in [11]. The following result shows that the two coincide whenthey both make sense. Theorem 3.3. If A is a cosemisimple Hopf algebra, then its cocenter HC ( A ) is a groupalgebra.Proof. According to the proof of [10, Proposition 2.4], a Hopf algebra map π : A → B iscocentral if and only if for any functional ϕ ∈ B ∗ the action of ϕ induced by π on any (right,say) A -comodule is an endomorphism in M A .We can now continue as in the proof in loc. cit. When A is cosemisimple the abovecondition is equivalent to saying that for every simple A -comodule V , the action ⊲ : B ∗ ⊗ V → V of B ∗ on V is of the form ϕ ⊲ v = ϕ ( g ) v, ∀ ϕ ∈ B ∗ , v ∈ V (3.1)for some grouplike element g ∈ B .If furthermore A → B is surjective (which is the case for B = HC ( A )), then B will begenerated by the grouplikes from (3.1). This proves the claim. (cid:4) In [10] an interpretation was given in the cosemisimple case for the group whose groupalgebra HC ( A ) is. In order to adapt that discussion to the present, general setting, we needthe following notions. First, we recall from [12, § Definition 3.4.
The blocks of an abelian category are the classes of the weakest equivalencerelation on the set of indecomposable objects generated by the requiring that two objects areequivalent if there are non-zero morphisms between them. (cid:7)
Definition 3.5.
The universal grading group Γ( A ) of a Hopf algebra A is the group with onegenerator g V for each simple V ∈ M A subject to the following relations • g V only depends on the block of the category M A of A -comodules that V belongs to; • g k = 1, where k ∈ M A denotes the trivial comodule; • For simple A -comodules V and W , the product g V g W equals g X for any simple sub-quotient X of V ⊗ W . (cid:7) Remark 3.6.
Note that it follows from the definition that g V ∗ = g − V , and similarly for theright dual ∗ V when the antipode is bijective (the bijectivity of the antipode is necessary inorder to define a coaction on ∗ V that makes it into a right dual). (cid:7) Remark 3.7.
The universal grading group is the chain group of [5, 18]. (cid:7)
We always have a Hopf algebra map A → k Γ( A ) from a Hopf algebra to the group algebraof its universal grading group: at the level of comodule categories, the corresponding functor M A → M k Γ( A ) sends an indecomposable A -comodule V to the direct sum of dim( V ) copiesof the k Γ( A )-comodule k g , where g is the element of Γ( A ) corresponding to the simple socleof V (recall that the socle of a comodule is the sum of its simple subcomodules).When A is cosemisimple A → k Γ( A ) can be identified with A → HC ( A ) (see [10]). Ingeneral, we again get a cocentral Hopf algebra quotient, smaller than HC ( A ). Definition 3.8.
The group cocenter of A is the quotient A → k Γ( A ) from the above discussion.We will sometimes also write HGC ( A ) for k Γ( A ). (cid:7) The following result shows that the group cocenter always fits into an exact sequence.
Theorem 3.9.
The group cocenter
HGC of A fits into an exact sequence k → C → A → HGC → k of Hopf algebras, with A faithfully flat over C and faithfully coflat over HGC .Proof.
Denote the surjection A → HGC by π and let C = { x ∈ A | ( π ⊗ id)(∆( x )) = 1 ⊗ x } . Since π is central, C is a Hopf algebra. Using [26, Theorem 2], we obtain faithful flatness of A over C given faithful coflatness of A over HGC ; this latter condition holds simply because π is a surjection onto a cosemisimple Hopf algebra, and hence automatically faithfully coflat(see Remark 1.3). (cid:4) Now let A be a Hopf algebra with bijective antipode and denote by HC and HGC its Hopfcocenter and group cocenter respectively. By the universality of A → HC , the surjection A → HGC factors as A HGCHC π (3.2)for a Hopf algebra surjection π . Note that π restricts to a surjection of the group G ( HC ) ofgrouplikes in HC onto Γ = Γ( A ) = G ( HGC ) (e.g. by [20, Proposition 4.1.7]). The next resultsays, essentially, that this surjection is in fact also one-to-one.
Proposition 3.10.
The surjection π from (3.2) is split as in the diagram HC HGC πι (3.3) by an inclusion ι of Hopf algebras.Proof. All we have to show is that, as in the discussion preceding the statement, π induces anisomorphism at the level of grouplikes. In other words, we need to prove that the grouplikesof HC satisfy the relations listed in Definition 3.5.As in the proof of Theorem 3.3, every grouplike element g ∈ G ( HC ) induces an action of HC ∗ on every A -comodule via the character ϕ ϕ ( g ) : HC ∗ → k . N THE HOPF (CO)CENTER OF A HOPF ALGEBRA 11
This character must be the same for comodules belonging to the same block because HC ∗ acts by A -comodule maps, hence the first bullet point in Definition 3.5.The second bullet point in that definition clearly holds for the group G ( HC ).Finally, the third bullet point of Definition 3.5 holds because HC is a Hopf algebra coactingon irreducible A -comodules via the grouplikes G ( HC ), and tensor products of comodulescorrespond to products of coalgebra coefficients with respect to the HC coaction. (cid:4) Hopf (co)center and adjoint coaction
Let A be a Hopf algebra and ad : A → A ⊗ A the adjoint coaction as defined in (1.4). Inthis section we shall describe the links between • the adjoint coaction ad and the Hopf center HZ ( A ), • the adjoint coaction ad and the Hopf cocenter HC ( A ).In what follows we shall provide necessary and sufficient conditions for ad : A → A ⊗ A to bean algebra homomorphism. Clearly, this is the case if A is commutative. Theorem 4.1.
Let A be a Hopf algebra and ad : A → A ⊗ A the adjoint coaction (1.4). Then ad is an algebra homomorphism if and only if ad ( A ) ⊂ A ⊗ Z ( A ) Proof.
The map ad is an algebra homomorphism if and only if its flipped version δ : A → A ⊗ A is where δ ( x ) = S ( x (1) ) x (3) ⊗ x (2) . In the course of the proof we shall use notation of the proofof [8, Theorem 3.3]. In particular we define the invertible linear map W : A ⊗ A → A ⊗ A : W ( a ⊗ a ′ ) = a (1) ⊗ a (2) a ′ It is easy to check that W W W − = W W (4.1)Let us consider a linear map U : A ⊗ A ⊗ A → A ⊗ A ⊗ A U ( a ⊗ b ⊗ c ) = (id ⊗ δ )(∆( a ))( ⊗ b ⊗ c ) = a (1) ⊗ S ( a (2) ) a (4) b ⊗ a (3) c for all a, b, c ∈ A . Then U satisfies W − U W = U U (4.2)For the proof of (4.2) see the proof of [8, Theorem 3.3]. Note that this is the place where theassumption that ad is an algebra homomorphism is used.Let us note that U = W − W W (4.3)Indeed W − W W ( a ⊗ b ⊗ c ) = W − W ( a (1) ⊗ a (2) b ⊗ c )= W − ( a (1) ⊗ a (3) b ⊗ a (2) c )= a (1) ⊗ S ( a (2) ) a (4) b ⊗ a (3) c Now, using (4.3) and (4.1) we get W − U W = W − W − W W W = W − W − W W W W = W − W − W W W W
232 ALEXANDRU CHIRVASITU AND PAWE L KASPRZAK
On the other hand, again using (4.3) we get U U = W − W W W − W W Comparing both expressions we conclude that W − W W W − = W − W − W W or equivalently W U = U W (4.4)Applying both sides of (4.4) to a ⊗ b ⊗ ⊗ ⊗ we get a (1) ⊗ b (1) ⊗ b (2) S ( a (2) ) a (4) ⊗ a (3) = a (1) ⊗ b (1) ⊗ S ( a (2) ) a (4) b (2) ⊗ a (3) (4.5)Applying ε ⊗ ε to the first two tensorands of both sides of (4.5) we get( b ⊗ )( S ( a (1) ) a (3) ⊗ a (2) ) = ( S ( a (1) ) a (3) ⊗ a (2) )( b ⊗ )i.e. ad ( A ) ⊂ A ⊗ Z ( A ).The converse implication is clear, i.e. if ad ( A ) ⊂ A ⊗ Z ( A ) then ad is an algebra homomor-phism and the theorem is proved. (cid:4) Assuming that ad ( A ) ⊂ A ⊗ Z ( A ), or equivalently that ad is an algebra homomorphism,and using Lemma 2.4 we get: Corollary 4.2.
Let A be Hopf algebra and HZ ( A ) its Hopf center. Then the adjoint coaction ad is an algebra homomorphism if and only if ad ( A ) ⊂ A ⊗ HZ ( A ) . In particular ad may beviewed as a coaction of HZ ( A ) on A . We find the following observation interesting in its own.
Lemma 4.3.
Let A be a Hopf algebra and x ∈ A . Then ad ( x ) = x ⊗ if and only if ∆( x ) = (id ⊗ S )(∆ op ( x )) Proof.
Assuming that x ⊗ = x (2) ⊗ S ( x (1) ) x (3) we get x (1) ⊗ x (2) ⊗ = x (2) ⊗ x (3) ⊗ S ( x (1) ) x (4) Thus x (2) ⊗ S ( x (1) ) = x (3) ⊗ S ( x (2) ) S ( x (1) ) x (4) = x (3) ⊗ S ( x (1) S ( x (2) )) x (4) = x (1) ⊗ x (2) Conversely, if x (1) ⊗ x (2) = x (2) ⊗ S ( x (1) )then x (1) ⊗ x (2) ⊗ x (3) = x (2) ⊗ x (3) ⊗ S ( x (1) )and we get x (2) ⊗ S ( x (1) ) x (3) = x (3) ⊗ S ( x (2) ) S ( x (1) )= x (3) ⊗ S ( S ( x (1) ) x (2) )= x (2) ⊗ ε ( x (1) ) = x ⊗ (cid:4) N THE HOPF (CO)CENTER OF A HOPF ALGEBRA 13
Let q : A → B be a Hopf morphism. Then, us proved in paragraph following [1, Definition2.3.8] q is a cocentral morphism if and only if( q ⊗ id)( ad ( x )) = q ( x ) ⊗ In what follows we provide an alternative characterization of cocentral Hopf morphisms.
Lemma 4.4.
Let q : A → B be a Hopf morphism. Then q is a cocentral if and only if (id ⊗ q )( ad ( x )) = x ⊗ Proof.
Let us consider a pair of morphisms π , π : A → A ⊗ B , π ( x ) = x ⊗ , π ( x ) = ⊗ q ( x ).It is easy to see that q is cocentral if and only if π ∗ π = π ∗ π . Defining a linear map π ( x ) = ⊗ S B ( q ( x )) = ⊗ q ( S A ( x )) we get π ∗ π ( x ) = π ( x (1) ) π ( x (2) )= ⊗ q ( x (1) ) S B ( q ( x ))= ⊗ q ( x (1) S A ( x (2) ))= ε ( x )( ⊗ )Similarly π ∗ π ( x ) = ε ( x )( ⊗ )Thus π is convolutively invertible. In particular q is cocentral if and only if π − ∗ π ∗ π = π i.e. if and only if x (2) ⊗ q ( S A ( x (1) ) x (3) ) = x ⊗ and we are done. (cid:4) Lemma 4.5.
Let us define C = linspan { ( ω ⊗ id)( ad ( x )) : ω ∈ A ∗ , x ∈ A } Then • C is a subcoalgebra of A and ad ( A ) ⊂ A ⊗ C ; • if q : A → B is cocentral then for all x ∈ C we have ( q ⊗ id)(∆( x )) = ⊗ x Proof.
Directly from the definition of C we get ad ( A ) ⊂ A ⊗ C . Using ad ( x ) ∈ A ⊗ C we get∆(( ω ⊗ id)( ad ( x ))) = ( ω ⊗ id ⊗ id)((id ⊗ ∆)( ad ( x )))= ( ω ⊗ id ⊗ id)(( ad ⊗ id)( ad ( x ))) ∈ ( ω ⊗ id)( ad ( A )) ⊗ C ⊂ C ⊗ C which shows that C is a coalgebra.For ω ∈ A ∗ and x ∈ A we get( q ⊗ id)(∆(( ω ⊗ id)( ad ( x )))) = ( ω ⊗ id ⊗ id)((id ⊗ q ⊗ id)(( ad ⊗ id)( ad ( x ))))= ( ω ⊗ id ⊗ id)( ad ( x ) ) = ⊗ ( ω ⊗ id)( ad ( x ))where in the second equality we used Lemma 4.4. (cid:4) Let D be the algebra generated by C . Since C is a coalgebra, D is a bialgebra: ∆( D ) ⊂ D ⊗ D .Let π : A → HC ( A ) be the canonical quotient. Using Lemma 4.5 we see that D ⊂ { x ∈ A : ( π ⊗ id)(∆( x )) = ⊗ x } We shall prove that D is preserved by a certain version of the adjoint action. Lemma 4.6.
Let A be a Hopf algebra with the bijective antipode S . For all x ∈ A and y ∈ D we have S ( x (1) ) yx (2) ∈ D Proof.
Let H denote the A -comodule whose underlying vector space is A and the comodulemap is ad . Let us view A as the A -comodule with the comodule map ∆. Let us considerthe comodules tensor products H ⊗ A and A ⊗ H . Using [10, Proposition 2.4] we see that T : H ⊗ A → A ⊗ H T ( h ⊗ x ) = x (1) ⊗ hx (2) yields the comodules identification where T − ( x ⊗ h ) = hS − ( x (1) ) ⊗ x (2) (4.6)In particular, denoting the corresponding comodules maps by ρ H ⊗ A and ρ A ⊗ H we have ρ H ⊗ A = ( T − ⊗ id) ◦ ρ A ⊗ H ◦ T In what follows m : A ⊗ A → A denotes the multiplication map and ω ∈ A ∗ a generic linearfunctional. Let us consider the element ω ( y (2) ) S ( y (1) ) y (3) ∈ C and x ∈ A . We compute ω ( y (2) ) S ( x (1) ) S ( y (1) ) y (3) x (2) = m (cid:0) (id ⊗ ω ⊗ ε ⊗ id)( S ( x (1) ) ⊗ y (2) ⊗ x (2) ⊗ S ( y (1) ) y (3) x (3) ) (cid:1) = m (cid:16) (id ⊗ ω ⊗ ε ⊗ id)( S ( x (1) ) ⊗ ρ H ⊗ A ( y ⊗ x (2) )) (cid:17) = m (cid:16) (id ⊗ (( ω ⊗ ε ) ◦ T − ) ⊗ id)( S ( x (1) ) ⊗ ρ A ⊗ H ( T ( y ⊗ x (2) ))) (cid:17) = m (cid:0) (id ⊗ (( ω ⊗ ε ) ◦ T − ) ⊗ id)( S ( x (1) ) ⊗ ρ A ⊗ H ( x (2) ⊗ yx (3) )) (cid:1) = m (cid:0) (id ⊗ (( ω ⊗ ε ) ◦ T − ) ⊗ id)( S ( x (1) ) ⊗ x (2) ⊗ (( ⊗ x (3) ) ad ( yx (4) )) (cid:1) Denoting Ω = ( ω ⊗ ε ) ◦ T − ∈ ( A ⊗ A ) ∗ and using (4.6) we getΩ( a ⊗ b ) = ( ω ⊗ ε )( aS − ( b (1) ) ⊗ b (2) ) = ω ( aS − ( b ))and ω ( y (2) ) S ( x (1) ) S ( y (1) ) y (3) x (2) = m (cid:0) (id ⊗ Ω ⊗ id)( S ( x (1) ) ⊗ x (2) ⊗ (( ⊗ x (3) ) ad ( yx (4) )) (cid:1) = m (cid:0) (id ⊗ Ω ⊗ id)( S ( x (1) ) x (3) ⊗ x (2) ⊗ ad ( yx (4) )) (cid:1) where in the last equality, moving x (3) towards S ( x (1) ), we used the fact that, at the endthe multiplication map m is applied. The last expression is of the form (here we use the legnumbering notation) m (cid:0) (id ⊗ Ω ⊗ id)( ad ( x (1) ) ad ( yx (2) ) ) (cid:1) Since ad ( A ) ⊂ A ⊗ C we get ad ( x (1) ) ad ( yx (2) ) ∈ C ⊗ A ⊗ A ⊗ C thus ω ( y (2) ) S ( x (1) ) S ( y (1) ) y (3) x (2) ∈ D (cid:4) N THE HOPF (CO)CENTER OF A HOPF ALGEBRA 15
Remark 4.7.
Following [22, § A with bijective antipode freelygenerated by the n × n matrix coalgebra C for some n ≥ b H ( C )).In this case, it can be shown that the subalgebra D defined above is not invariant under theantipode.Hence, it is possible for D to be an ad-invariant sub-bialgebra of A but not a Hopf subal-gebra. (cid:7) Cocentral exact sequence
The following result will be crucial for the proof that a Hopf algebra and its cocenter fitinto an exact sequence in the sense of Definition 1.1. Before we state it, we need the followingnotion.
Definition 5.1.
Let f : A → B be a Hopf algebra morphism. We say that f trivializes asubset S ⊂ A (or that S is trivialized by f ) if((id ⊗ f ) ◦ ∆)( s ) = s ⊗ , ∀ s ∈ S . Under the above setup, the
Hopf reflection of S ⊂ A is a Hopf algebra morphism A → B that trivializes S and is universal with this property, in the following sense:any Hopf algebra map A → B ′ that trivializes S factors uniquely as A B B ′ A Hopf algebra map A → B is saturated if it is the Hopf reflection of some subset S ⊆ A . (cid:7) Remark 5.2.
Note that the Hopf reflection of a subset
S ⊆ A may or may not exist, but ifit does it is a uniquely determined Hopf quotient of A . (cid:7) With this in place, we can state the result alluded to above.
Theorem 5.3.
Every saturated Hopf algebra map f : A → B invariant under either the leftor right adjoint coaction of A on itself fits into an exact sequence k → C → A → B → k of Hopf algebras in the sense of Definition 1.1.Proof. Let
S ⊆ A be a subset whose Hopf reflection f is. Define C ⊂ A by C = { a ∈ A | ((id ⊗ f ) ◦ ∆)( a ) = a ⊗ ∈ A ⊗ B } . The adjoint coaction invariance assumption on f ensures that C is a Hopf algebra (by thedual version of [2, Lemma 1.1.11]). Moreover, we then know from [2, Lemma 1.1.12 (i)] that A → A / AC + is a Hopf algebra quotient. According to Definition 1.1 we will be done with theproof of the theorem once we argue that AC + = ker( f ) (i.e. A → A / AC + can be identifiedwith f ).Since A → A / AC + is the Hopf cokernel of C → A which in turn is the Hopf kernel of f , wealready know that the latter map factors through A / AC + , and hence AC + ⊆ ker( f ).On the other hand C contains S , so the universality property implicit in the saturationhypothesis shows that we have a factorization A B A / AC + f thus getting the reverse inclusion ker( f ) ⊆ AC + . (cid:4) As a consequence, we get
Corollary 5.4.
Let A be an arbitrary Hopf algebra. Then, the cocenter HC = HC ( A ) fits intoan exact sequence k → C → A → HC → k (5.1) of Hopf algebras.Proof. A → HC is invariant under both adjoint coactions, and it is saturated by Lemmas 4.4and 4.5 with S being the set of right hand tensorands of x (2) ⊗ S ( x (1) ) x (3) ∈ A ⊗ A for x ∈ A . Hence Theorem 5.3 applies to this situation and we are done. (cid:4) Exact sequences are particularly pleasant when we know in addition that the Hopf algebrain the middle is faithfully (co)flat over the other terms. We have the following analogue ofTheorem 3.9 for the cocenter (rather than the group cocenter) in case A is pointed. Corollary 5.5.
Let A be a pointed Hopf algebra. Then, the cocenter HC = HC ( A ) fits intoan exact sequence k → C → A → HC → k of Hopf algebras, with A faithfully flat over C and faithfully coflat over HC .Proof. The claim on the existence of the exact sequence is Corollary 5.4, so only the (co)flatnessclaims need proof.According to [20, Theorem 9.3.1] (or [19]), pointed Hopf algebras are faithfully flat (free,in fact) over their Hopf subalgebras. Hence we know that A is faithfully flat over C , so thataccording to [26, Theorem 1] in the exact sequence k → C → A → A / AC + → kA faithfully coflat over A / AC + and H ∼ = AC + . (cid:4) In fact, with a little more work we can extend this result to the fully general setup of thispaper.
Theorem 5.6.
Every Hopf algebra A is faithfully flat over its cocenter HC , and faithfully flatover the Hopf kernel C of the cocenter. Before going into the proof, we need some preparation. The context for the next result isas follows. • A is an arbitrary Hopf algebra; • C ′ ⊆ A is the Hopf kernel of the group cocenter A → HGC (i.e. the left hand term ofthe exact sequence in Theorem 3.9); • C ⊆ C ′ is the Hopf kernel of the cocenter A → HC . Lemma 5.7.
With the above notation the Hopf algebras C ⊆ C ′ share the same coradical.Proof. Let V be a simple A -comodule. When regarded as an HGC -comodule, V breaks up as adirect sum of copies of the same one-dimensional comodule corresponding to some grouplikein the universal grading group Γ( A ) = G ( HGC ) (see Theorem 3.9 and surrounding discussion).On the other hand, since V ∈ M A is simple and the dual HC ∗ acts on A -comodules via A -comodule morphisms, the same reasoning as in the proof of Theorem 3.3 shows that V is N THE HOPF (CO)CENTER OF A HOPF ALGEBRA 17 also a sum of copies of the same one-dimensional HC -comodule. Moreover, we know fromProposition 3.10 that HC → HGC induces an isomorphism at the level of coradicals, so HC coacts trivially on V if and only if HGC does.We can now finish the proof: the coradical of C ′ ( C ) consists of those simple subcoalgebrasof A that are coefficient subcoalgebras for the simple A -comodules coacted upon trivially by HGC (respectively HC ). (cid:4) Proof of Theorem 5.6.
Since we already have an exact sequence (5.1), the flatness and coflat-ness assertions are equivalent to one another by [26, Theorems 1 and 2]. Hence, it suffices toshow that A is flat over the Hopf kernel C of the cocenter.We know from Theorem 3.9 that A is faithfully flat over the Hopf kernel C ′ of the groupcocenter A → HGC .On the other hand, C ⊆ C ′ is an inclusion of Hopf algebras with a common coradical.According to [19, Corollary 1], C ′ is faithfully flat (and in fact free) over C . The desiredconclusion follows from the transitivity of faithful flatness. (cid:4) We now turn to the issue of identifying the Hopf kernel C of A → HC more explicitly. Thefollowing general result in the context of saturated Hopf algebra morphisms will be of use inthat respect. Lemma 5.8.
Let f : A → B be a Hopf algebra map satisfying the hypotheses of Theorem 5.3which is the Hopf reflection of some subset S ⊆ A .Let D ⊆ A be a Hopf subalgebra such that (a) D contains S , (b) D is invariant under either the left or the right adjoint action of A on itself and (c) A is left faithfully flat over D .Then, D contains C . Moreover, if D is generated by S as a Hopf algebra, then D = C .Proof. Conditions (a) and (b) ensure that A → A / AD + is a quotient Hopf algebra that trivial-izes S . By the universality of A → B implicit in being the reflection of S (see Definition 5.1),we have a factorization A B A / AD + f (5.2)The faithful flatness hypothesis implies via [26, Theorem 1] that D is precisely the Hopf kernelof A → A / AD + . Since C is the Hopf kernel of A → B , the conclusion C ⊆ D follows fromcommutative diagram (5.2).As for the last statement, if D is generated by S and C ⊇ S is another Hopf algebracontaining S , the opposite inclusion C ⊇ D also holds. (cid:4) We will now seek to apply Lemma 5.8 to the sub-bialgebra D ⊆ A defined in the discussionfollowing Lemma 4.5 (generated by the right hand tensorands of the right adjoint coaction(1.4)). Taking S ⊆ A to be the subspace of right hand tensorands of x (2) ⊗ S ( x (1) ) x (3) x ∈ A , we get Corollary 5.9.
Let A be a Hopf algebra, and suppose D ⊆ A defined as above is a Hopfalgebra and A is left faithfully flat over D . Then, D is the Hopf kernel of the cocenter of A . Proof.
This is an immediate consequence of Lemma 5.8. (cid:4)
Remark 5.10. If D is not a Hopf algebra then let D ′ be the smallest Hopf subalgebra of A containing D and being preserved by the adjoint action. Then assuming that A is faithfullyflat over D ′ and reasoning as in the proof of Corollary 5.9 we get the exact sequence k → D ′ → A → HC → k (cid:7) Example 5.11.
Let A be a cosemisimple Hopf algebra and [ u βij ] = u β a finite dimensionalirreducible corepresentation of A . Then ad ( u βij ) = X k,l u βkl ⊗ S ( u βik ) u βlj Thus D is generated by the elements S ( u βik ) u βlj where β runs over the equivalence classes ofirreducible corepresentations of A and i, j, k, l ∈ { , . . . , dim u β } . Noting that S ( S ( u βik ) u βlj ) = S ( u βlj ) S ( u βik )and using the fact that [ S ( u βik )] is in the discussed cosemisimple case equivalent with u β wesee that the set generating D is preserved by S . Thus D is a Hopf subalgebra of A which ispreserved by the adjoint action (see Lemma 4.6) and since A is faithfully flat over D (see [11])then by Corollary 5.9 we get the exact sequence k → D → A → HC → k Summarizing we get an alternative proof of [11, Proposition 2.13]. (cid:7)
We will see below that the faithful flatness assumption in Corollary 5.9 can essentially bedropped. In order to do this, we need some preliminaries. We will be referring to [20] forbackground on comodule theory. Specifically, we recollect some notions from [20, Sections3.7, 4.8].
Definition 5.12.
Two simple subcoalgebras C , D of a coalgebra A are directly linked if thespace C ∧ D + D ∧ C ⊆ A is strictly larger than C + D . C and D are linked if there is a finite sequence of simple subcoalgebras starting with C andending with D , so that every two consecutive ones are directly linked.A coalgebra is link indecomposable if all of its simple subcoalgebras are linked.A link indecomposable component (or just ‘component’ for short) of a coalgebra is a maximallink indecomposable subcoalgebra. (cid:7) Remark 5.13. If V and W are the simple comodules corresponding to the simple subcoal-gebras C and D respectively, then C and D are directly linked if and only if there is anindecomposable A -comodule admitting a composition series whose components are V and W .In other words, there is a non-split exact sequence0 → V → • → W → → W → • → V → M A . More precisely, (5.3) corresponds to C ⊕ D ( C ∧ D while the existence of a non-splitsequence (5.4) is equivalent to C ⊕ D ( D ∧ C . N THE HOPF (CO)CENTER OF A HOPF ALGEBRA 19
Every coalgebra breaks up as the direct sum of its components. In other words, thecomponents of a coalgebra A correspond to the blocks of the category M A . (cid:7) We now introduce the following notion.
Definition 5.14.
Let A be a coalgebra. A subcoalgebra C ⊆ A is coradically replete if forany two linked simple subcoalgebras D , E ∈ A the inclusion D ⊆ C implies E ⊆ C . C is replete if it is a direct summand of A as a coalgebra (i.e. C is a direct sum of some ofthe components of A ). (cid:7) We are now ready to state a number of auxiliary results.
Lemma 5.15.
Let A be a Hopf algebra and C ⊆ A a replete Hopf subalgebra. Then, theinclusion C → A splits as a map of C -bimodules and A -bicomodules.Proof. By hypothesis, we have a decomposition A = C ⊕ D as coalgebras. We will be done ifwe argue that D is a C -bimodule, since in that case the required splitting will simply be theprojection A = C ⊕ D → C .Our task amounts to showing that if V and W are comodules over C and D respectively,then their tensor product in M A does not contain any C -subcomodules. Since comodulesare unions of their finite-dimensional subcomodules, we will assume that V and W are finite-dimensional.Suppose we have some non-zero map U → V ⊗ W for U ∈ M C . This means we have anon-zero map V ∗ ⊗ U → W , contradicting the fact that V ∗ ⊗ U is a C -comodule (because thelatter is a Hopf subalgebra of A ).On the other hand, a non-zero map U → W ⊗ V would provide a non-zero morphism W ∗ → V ⊗ U ∗ , and we can repeat the argument. (cid:4) As a consequence, we get
Proposition 5.16.
A Hopf algebra is left and right faithfully flat over any replete Hopfsubalgebra.Proof.
The splitting in Lemma 5.15 implies faithful flatness as in the proof of the main resultin [11]. (cid:4)
We can strengthen this as follows.
Theorem 5.17.
A Hopf algebra is left and right faithfully flat over any coradically repleteHopf subalgebra.Proof.
Let A be a Hopf algebra and C ⊆ A a coradically replete Hopf subalgebra.Consider the subcoalgebra D ⊆ A defined as the sum of all components of A that intersect C non-trivially. D is a Hopf subalgebra, and C ⊆ D ⊆ A . As in the proof of Theorem 5.6, theleft hand inclusion is faithfully flat by [19, Corollary 1] because the two Hopf algebras sharethe same coradical. On the other hand, D ⊆ A is faithfully flat by Proposition 5.16. (cid:4) Theorem 5.17 is relevant to us because of the following
Lemma 5.18.
Let A be a Hopf algebra with bijective antipode. If the sub-bialgebra D ⊆ A generated by the right hand tensorands of (1.4) is a Hopf subalgebra with bijective antipode,then it is coradically replete. Proof.
We need to show that if C ⊆ D and E ⊆ A are directly linked simple subcoalgebras,then E too is contained in D .According to the definition of being directly linked (see Definition 5.12), there are twopossibilities we have to treat. Case 1: C ⊕ E ( C ∧ E . Let V and W be the simple right comodules over C and E respectively entering (5.3). Our hypotheses then ensure via Lemma 5.19 below that V ∗ ⊗ W isa D -comodule. Since V is also a D -comodule by assumption and W is a quotient of V ⊗ V ∗ ⊗ W ,we have W ∈ M D and hence E ⊆ D . Case 2: C ⊕ E ( E ∧ C . Keeping the above notation for V and W , this time around weknow that W ∗ ⊗ V is a D -comodule, along with V . This means first that W ∗ , being a quotientof W ∗ ⊗ V ⊗ V ∗ , must be a D -comodule. In turn, this implies W ∈ M D because D is preservedby the inverse of the antipode (this is where that hypothesis is necessary). (cid:4) Lemma 5.19.
Let C ⊆ A be the subcoalgebra spanned by the right hand tensorands of theadjoint coaction (1.4). Suppose V and W are simple A -comodules for which there is a non-splitexact sequence → V → U → W → in M A . Then, V ∗ ⊗ W is a C -comodule.Proof. Consider the canonical coalgebra map φ : U ∗ ⊗ U → A implementing the A -coactionon U . The fact that (5.5) does not split translates to the fact that φ factors through anembedding into A of the quotient coalgebra E = (cid:18) V ∗ ⊗ V V ∗ ⊗ W W ∗ ⊗ W (cid:19) (5.6)of U ∗ ⊗ U (in other words, its quotient by W ∗ ⊗ V ).If we think of U ∗ ⊗ U as an object of M A with the tensor product comodule structure,then it is a coalgebra in the monoidal category M A , and its coaction on U is a morphismof coalgebras. Additionally, the map φ is one of A -comodules if we give A its right adjointcoaction, and so E is an A -comodule coalgebra coacting on U so that the comodule structuremap U → U ⊗ E is one of A -comodules. As x ranges over the image E of φ , the right handtensorands of the adjoint coaction (1.4) span the coefficient coalgebra of U ∗ ⊗ U ∈ M A . Inconclusion, E ∈ M A is actually a C -comodule.As the subcategory M C ⊆ M A is closed under taking subobjects (it is closed in the sense ofDefinition 2.7), the A -comodule coalgebras V ∗ ⊗ V ⊆ E and W ∗ ⊗ W ⊆ E are both C -comodules.In conclusion, so is the quotient V ∗ ⊗ W of E by their sum. (cid:4) Remark 5.20.
Note that when A has bijective antipode, the antipode bijectivity of D isnecessary for faithful flatness (e.g. by [9, Theorem 3.2 (b)]); Lemma 5.18 reverses this impli-cation. (cid:7) Finally from Lemma 5.18 and Theorem 5.17, we get
Corollary 5.21.
Under the assumptions of Lemma 5.18 A is left and right faithfully flat over D . (cid:4) Examples
The center and cocenter of q -deformations of G . Let G be a compact semisimplesimply connected Lie group, G q the Drinfeld-Jimbo quantization of G for some transcendental q ∈ C × , and A = Pol( G q ). Using [15, Theorem 9.3.20] we see that Z ( A ) = C thus HZ ( A ) = C . N THE HOPF (CO)CENTER OF A HOPF ALGEBRA 21
Since the cocenter of a semisimple Hopf algebra Pol( G q ) depends only on the fusion ringand the fusion ring of Pol( G q ) is the same as the fusion ring of Pol( G ) then we see that HC (Pol( G q )) = HC (Pol( G )). The later is the group algebra of the center of G . Remark 6.1.
The situation is more complicated when q is a root of unity. In that case, theappropriately-defined Pol( G q ) contains a central Hopf subalgebra isomorphic to Pol( G ), andthe resulting exact sequences can be applied to the study and classification of finite quantumsubgroups of G q ; see e.g. [13, 3]. (cid:7) Drinfeld twist.
Let A be a Hopf algebra and Ψ ∈ A ⊗ A an invertible 2-cocycle. Let∆ Ψ : A → A ⊗ A be the twisted comultiplication∆ Ψ ( x ) = Ψ∆( x )Ψ − It is well known that A equipped with ∆ Ψ is a Hopf algebra which we denote by A Ψ . UsingLemma 2.3 we get HZ ( A Ψ ) = HZ ( A ).6.3. Center of U q ( g ) . Let us consider the quantized enveloping algebra U q ( g ), defined, say,as in [17, § q ∈ C × and g is asimple complex Lie algebra of rank ℓ .Let α i , 1 ≤ i ≤ ℓ be a set of simple roots for g and ( a ij ) the associated Cartan matrix.Finally, we set d i = ( α i , α i ) / α i and define q i = q d i . U q ( g ) then has generators E i , F i , K i , ≤ i ≤ ℓ such that K i are invertible and commute with one another, K i E j K − i = q a ij i E j , K i F j K − i = q − a ij i F j , [ E i , F j ] = δ ij K i − K − i q i − q − i (6.1)(where we are assuming q d i = 1) and some additional relations are satisfied that are meantto mimic the Serre relations for g . We refer to loc. cit. for details.The algebra U q ( g ) has a unique Hopf algebra structure making K i grouplike and for which∆( E i ) = E i ⊗ K i + 1 ⊗ E i , ∆( F i ) = F i ⊗ K − i ⊗ F i . (6.2) Proposition 6.2. If q is not a root of unity then the Hopf center HZ = HZ ( U q ( g )) is C .Proof. Note that U = U q ( g ) is pointed as a coalgebra ([23, p. 157]) in the sense that its simplecomodules are one-dimensional (they correspond to the grouplikes, which make up the groupgenerated by the K i ). Since HZ is a Hopf subalgebra of U , it too must be pointed. Claim 1: HZ is irreducible as a coalgebra, i.e. is its only grouplike. Indeed,the set of grouplikes is the group h K i , ≤ i ≤ ℓ i ∼ = Z ℓ . A non-trivial element of it is of the form g = Q i K n i i with at least one non-zero n i . Fixingsuch an i , the conjugate gE i g − is q ( λ,α i ) E i , where λ = P n i α i . Now, ( λ, α i ) cannot bezero for all i unless λ itself is zero, because of the non-degeneracy of ( − , − ). In addition, if( λ, α i ) = 0, then gE i g − = q ( λ,α i ) E i = E i because q is not a root of unity. In conclusion, the centrality of g implies n i = 0 for all i .This proves the claim. Now suppose HZ is not trivial. Then, since it is pointed irreducible, it must containnon-zero primitive elements, i.e. x ∈ HZ such that∆( x ) = x ⊗ ⊗ x (e.g. [23, Corollary 11.0.2]). We will have reached the desired contradiction once we prove Claim 2: U has no non-zero central primitive elements. Let 0 = x ∈ HZ be aprimitive element. Being central, x acts as a scalar c λ on each simple U -module of type V λ corresponding to some dominant weight λ of g . Since U acts jointly faithfully on all V λ , x = 0implies that c λ = 0 for at least some λ .Recall e.g. from [17, § U has a PBW basis, in the sense that the elements F r β · · · F r n β n K t · · · K t ℓ ℓ E s n β n · · · E s β (6.3)form a vector space basis. Here, the r i and s i are non-negative integers, the t i are integers,and β up to β n are the positive roots of g ordered in a certain way; see loc. cit. for details.Now, x is a linear combination of elements of the form (6.3). Under this expansion, thosesummands that contain at least some E s annihilate a highest weight vector v λ for every V λ .Since there are at least some λ for which v λ is not annihilated, we must have summands in x that contain no E s. But these summands can then contain no F s either, because elements ofthe form (6.3) with F s but no E s will strictly lower weights, whereas x , being central, cannot.In conclusion, when expanded according to the basis (6.3) x contains summands of theform K t · · · K t ℓ ℓ , (6.4)while all other summands contain both E s and F s.Now let x res be the sum of all summands (6.4) of x , with the respective coefficients (so x res is obtained from x by simply dropping the summands containing E s and F s).Now, both the algebra U + generated by the K s and E s and the algebra U − generated by K s and F s are Hopf subalgebras of U . They are graded as Hopf algebras viadeg( K ) = 0 , deg( E ) = 1 for U + , deg( K ) = 0 , deg( F ) = 1 for U − This means that when applying the coproduct to the summands in x − x res and expand withrespect to the tensor product of PBW basis of U ⊗ U , we obtain only terms that contain E sand F s. In conclusion, if U ⊂ U is the algebra generated by the K s, then the only summandsof ∆( x ) = x ⊗ ⊗ x that belong to U ⊗ U come from ∆( x res ).It follows from the discussion above that x res itself must be a primitive element. This isimpossible, since it belongs to the group algebra U which has no non-zero primitives. (cid:4) The next result complements Proposition 6.2 for (some) roots of unity, but we need apreparation before stating it. In both the discussion preceding and the proof of Proposition 6.3below we will follow [17, Section 6.3.5].Let q ∈ C be a primitive p th root of unity for some odd integer p > d i , i = 1 , , . . . , l . For aroot α we consider E α , F α , as in the proof of Proposition 6.2. Then [17, Proposition 47] saysthat e α := E pα , f α := F pα and k i = K pi are central in U q = U q ( g ). For the simple root α i we shall also write e i = E pα i , f i = F pα i .Then, according to [17, Proposition 48] the algebra ζ generated by e α , f α and k i is a Hopfsubalgebra of U . N THE HOPF (CO)CENTER OF A HOPF ALGEBRA 23
In particular, the Hopf center of U q is non-trivial and it contains ζ . With this in place,we can now state Proposition 6.3.
When q ∈ C × is a root of unity whose order p is odd and larger than all d i , the Hopf center of U q is the algebra ζ defined above.Proof. Consider the inclusion ζ ⊆ HZ = HZ ( U q ) of commutative Hopf algebras.We know from [17, Theorem 4.9] that U q is a finitely-generated free module over ζ , andhence HZ is finitely generated as a module over ζ (e.g. because the latter is a finitelygenerated commutative ring, and hence noetherian). It follows from this that in the exactsequence C → ζ → HZ → • → C of Hopf algebras the third term • is finite-dimensional. Since it is also commutative and weare working in characteristic zero, it must be the function algebra C ( G ) of some finite group G . It is easy to see that the central grouplikes in U q are precisely those in the group generatedby the k i = K pi , and hence the inclusion ζ ⊆ HZ is an isomorphism at the level of coradicals.But this means that the cosemisimple Hopf algebra C ( G ) is pointed irreducible, meaning that G is trivial. (cid:4) Remark 6.4.
It seems likely that the determination of the centers of two-parameter quantumgroups U r,s ( sl n ) carried out in [6] can be put to similar use, though we will not attempt thishere. (cid:7) Cocenter of U q ( g ) . In what follows we denote C the coalgebra assigned to U q ( g ) asdescribed in general in Lemma 4.5; D is the bialgebra generated by C . Lemma 6.5.
Let π : U q ( g ) → B be a cocentral morphism of Hopf algebras. Then π ( x ) = ε ( x ) . In particular HC ( U q ( g )) = C .Proof. The cocentrality of π yields π ( E i ) ⊗ K i + ⊗ E i = π ( K i ) ⊗ E i + π ( E i ) ⊗ Now the linear independence of K i , , E i implies that π ( E i ) = 0 and π ( K i ) = . Similarlywe are reasoning for F i = 0 and we are done. (cid:4) The above result shows that the Hopf kernel assigned to the surjection U q ( g ) → HC ( U q ( g ))is equal U q ( g ). We shall show that already D = U q ( g ). Computing in what follows we shalldrop the index i : (∆ ⊗ id)(∆( E )) = E ⊗ K ⊗ K + ⊗ E ⊗ K + ⊗ ⊗ E Thus ad ( E ) = − K ⊗ E + E ⊗ K + ⊗ E (6.5)The linear independence of E, K, enables us to slice the first leg of (6.5) by functionalswhich put K to 1 and E, to zero. Thus we get E ∈ D . Similarly K ∈ D . The same appliedto F yields F ∈ D . Summarizing we proved Lemma 6.6.
Adopting the above notation we have D = U q ( g ) . General concluding remarks.
The following simple observation can be useful in com-puting Hopf cocenters. Before stating it, we need the following
Definition 6.7.
The cocenter of a coalgebra C is the finite dual of the center of C ∗ .We denote the cocenter of C by CZ ( C ). (cid:7) Remark 6.8.
Note that we have a surjection C → CZ ( C ), universal among morphisms ofcoalgebras defined on C that are cocentral in the same sense as in Definition 3.1. (cid:7) Lemma 6.9. (a)
If a Hopf algebra A is generated by a subcoalgebra C as a Hopf algebra,then the kernel of A → HC ( A ) is the Hopf ideal generated by the kernel of C → CZ ( C ) . (b) The analogous statement holds for Hopf algebras with bijective antipode.Proof.
The difference between parts (a) and (b) is that in the former case A is generated asan algebra by the iterations S n ( C ) for n ∈ Z ≥ , whereas in the latter we allow n to range overall integers. To fix ideas, we focus on part (a); the proof for (b) is completely analogous.By the universality of C → CZ we have a commutative diagram C A
CZ HC of coalgebra maps, and hence the kernel of C → CZ is indeed contained in that of A → HC .On the other hand, consider the Hopf ideal I in the statement (generated by ker( C → CZ )).The quotient A → A / I is cocentral in the sense of Definition 3.1 because C → A / I is, theiterations S n ( C ) generate A as an algebra, and the condition from Definition 3.1 is preservedunder taking products and antipodes (i.e. if it holds for x, y ∈ A it also holds for xy , etc.).It follows from the previous paragraph that A → A / I factors through A → HC , and henceker( A → HC ) ⊆ I . Since the preceding discussion argues that the opposite inclusion holds, we are done. (cid:4)
Remark 6.10.
Note that Lemma 6.5 can be read as a particular instance of Lemma 6.9. (cid:7)
Lemma 6.9 allows us to compute the cocenters of Hopf algebras freely generated by coalge-bras in the sense of [24]. We omit the proof of the following result, as it is almost tautological.
Corollary 6.11.
The Hopf cocenter of the Hopf algebra H ( C ) freely generated by a coalgebra C is the free Hopf algebra H ( CZ ) on the cocenter CZ = CZ ( C ) . (cid:4) References [1] N. Andruskiewitsch. Notes on extensions of Hopf algebras.
Can. J. Math. , 48(1):3–42, 1996.[2] N. Andruskiewitsch and J. Devoto. Extensions of Hopf algebras.
Algebra i Analiz , 7(1):22–61, 1995.[3] N. Andruskiewitsch and G. A. Garc´ıa. Extensions of finite quantum groups by finite groups.
Transform.Groups , 14(1):1–27, 2009.[4] Sergey Arkhipov and Dennis Gaitsgory. Another realization of the category of modules over the smallquantum group.
Adv. Math. , 173(1):114–143, 2003.[5] Hellmut Baumg¨artel and Fernando Lled´o. Duality of compact groups and Hilbert C ∗ -systems for C ∗ -algebras with a nontrivial center. Internat. J. Math. , 15(8):759–812, 2004.[6] Georgia Benkart, Seok-Jin Kang, and Kyu-Hwan Lee. On the centre of two-parameter quantum groups.
Proc. Roy. Soc. Edinburgh Sect. A , 136(3):445–472, 2006.
N THE HOPF (CO)CENTER OF A HOPF ALGEBRA 25 [7] Tomasz Brzezi´nski and Robert Wisbauer.
Corings and comodules , volume 309 of
London MathematicalSociety Lecture Note Series . Cambridge University Press, Cambridge, 2003.[8] M. Budzi´nski and P. Kasprzak. On the quantum families of quantum group homomorphisms.
Accepted toCommun. Algebra .[9] Alexandru Chirv˘asitu. On epimorphisms and monomorphisms of Hopf algebras.
J. Algebra , 323(5):1593–1606, 2010.[10] Alexandru Chirv˘asitu. Centers, cocenters and simple quantum groups.
J. Pure Appl. Algebra , 218(8):1418–1430, 2014.[11] Alexandru Chirv˘asitu. Cosemisimple Hopf algebras are faithfully flat over Hopf subalgebras.
AlgebraNumber Theory , 8(5):1179–1199, 2014.[12] Jonathan Comes and Victor Ostrik. On blocks of Deligne’s category Rep( S t ). Adv. Math. , 226(2):1331–1377, 2011.[13] Corrado De Concini and Volodimir Lyubashenko. Quantum function algebra at roots of 1.
Adv. Math. ,108(2):205–262, 1994.[14] Heinz Hopf. ¨Uber die Topologie der Gruppen-Mannigfaltigkeiten und ihre Verallgemeinerungen.
Ann. ofMath. , 42:22–52, 1941.[15] A Joseph.
Quantum groups and their primitive ideals . Springer.[16] P. Kasprzak, A. Skalski, and P.M. So ltan. Short exact sequence { e } → Z ( G ) → G → Inn( G ) → { e } forlocally compact quantum groups. arXiv:1508.02943 preprint .[17] Anatoli Klimyk and Konrad Schm¨udgen. Quantum groups and their representations . Texts and Mono-graphs in Physics. Springer-Verlag, Berlin, 1997.[18] Michael M¨uger. On the center of a compact group.
Int. Math. Res. Not. , (51):2751–2756, 2004.[19] David E. Radford. Pointed Hopf algebras are free over Hopf subalgebras.
J. Algebra , 45(2):266–273, 1977.[20] David E. Radford.
Hopf algebras , volume 49 of
Series on Knots and Everything . World Scientific PublishingCo. Pte. Ltd., Hackensack, NJ, 2012.[21] Peter Schauenburg.
Tannaka duality for arbitrary Hopf algebras , volume 66 of
Algebra Berichte [AlgebraReports] . Verlag Reinhard Fischer, Munich, 1992.[22] Peter Schauenburg. Faithful flatness over Hopf subalgebras: counterexamples. In
Interactions between ringtheory and representations of algebras (Murcia) , volume 210 of
Lecture Notes in Pure and Appl. Math. ,pages 331–344. Dekker, New York, 2000.[23] Moss E. Sweedler.
Hopf algebras . Mathematics Lecture Note Series. W. A. Benjamin, Inc., New York,1969.[24] Mitsuhiro Takeuchi. Free Hopf algebras generated by coalgebras.
J. Math. Soc. Japan , 23:561–582, 1971.[25] Mitsuhiro Takeuchi. Formal schemes over fields.
Comm. Algebra , 5(14):1483–1528, 1977.[26] Mitsuhiro Takeuchi. Relative Hopf modules—equivalences and freeness criteria.
J. Algebra , 60(2):452–471,1979.
Department of Mathematics, University of Washington, Seattle, WA, USA
E-mail address : [email protected] Department of Mathematical Methods in Physics, Faculty of Physics, University of Warsaw,Poland
E-mail address ::